Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Math 728 - Homework # 1 - Peter Lewis Exercise 4.1.2. The weights of 26 professional baseball pitchers are given below. Suppose we assume that the weight of a professional baseball pitcher is normally distributed with mean µ and variance σ 2 . 160 175 180 185 185 185 190 190 195 195 195 200 200 200 200 205 205 210 210 218 219 220 222 225 225 232 (a) Obtain a frequency distribution and a histogram or a stem-leaf plot of the data. Use 5-pound intervals. Based on this plot, is a normal probability model credible? Solution: Based on the above data, I have constructed the following histogram, in which the intervals of weights are of the form [a, a + 5). Weights of Baseball Players Frequency 4 3 2 1 0 160 165 170 175 180 185 190 195 200 205 210 215 220 225 230 235 Based on this plot, a normal probability model does not seem credible. (b) Obtain the maximum likelihood estimators of µ, σ 2 , σ, and µ/σ. Locate your estimate of µ on your plot in part (a). Solution: Let f (x; µ, σ) be the pdf of the normal population. Then the likelihood function is L(µ, σ; x) = 26 Y i=1 26 Y f (xi ; µ, σ) (xi − µ)2 √ = exp − 2σ 2 2πσ 2 i=1 ( ) 26 X 1 = (2πσ 2 )−13 exp − 2 (xi − µ)2 . 2σ i=1 1 1 So, log likelihood follows as 26 1 X (xi − µ)2 . l(µ, σ; x) = ln L(µ, σ; x) = −13ln(2πσ ) − 2 2σ i=1 2 • Now, to find the MLE of µ, we have 26 ∂l(µ, σ; x) 1 X = 2 (xi − µ), ∂µ 2σ i=1 and 26 ∂l(µ, σ; x) 1 X = 0 =⇒ µ = xi ≈ 200.6. ∂µ 26 i=1 Since 13 ∂ 2 l(µ, σ; x) = − 2 < 0, 2 ∂µ σ x̄ maximizes l, and hence µ̂ = X̄ is the MLE of µ. • For the MLE of σ 2 , we have 26 26 ∂l(µ, σ; x) 13 1 X 1 X 2 2 = − + (x − µ) = 0 =⇒ σ = (xi − µ)2 ≈ 293.92, i ∂σ 2 σ 2 2σ 4 i=1 26 i=1 and " # 26 26 X 13 1 X 1 ∂ 2 l(µ, σ; x) 1 = 4− 6 (xi − µ)2 = 4 13 − 2 (xi − µ)2 . ∂(σ 2 )2 σ σ i=1 σ σ i=1 If σ 2 = 1 26 P26 i=1 (xi − µ)2 , we have 26 1 X 13 − 2 (xi − µ)2 = 13 − 26 < 0. σ i=1 Hence, σˆ2 = 1 26 P26 i=1 (Xi − µ)2 is the MLE of σ 2 . • By the invariance property of MLEs, we have σ̂ = the MLE of σ. q 1 26 • The MLE of µ/σ is approximately 200.6/17.14 ≈ 11.7 2 P26 i=1 (Xi − µ)2 ≈ 17.14 is (c) Using the binomial model, obtain the maximum likelihood estimate of the proportion p of professional baseball pitchers who weigh over 215 pounds. Solution: Notice that there are 7 players above 215 pounds. If we assume a binomial model, our likelihood function is just the pmf of Binomial(26,p): 26 x L= p (1 − p)26−x . x We ignore the 26 constant, as we’re maximizing in p. So, differentiate in p: x xpx−1 (1 − p)26−x − (26 − x)(1 − p)25−x px = px−1 (1 − p)25−x [x(1 − p) − (26 − x)p] = px−1 (1 − p)25−x [x − 26p], and equating to zero yields p = x/26. =⇒ p̂ = 7/26 is the MLE of p. (d) Determine the mle of p assuming that the weight of a professional baseball player follows the normal probability model N (µ, σ 2 ) with µ and σ unknown. Solution: Using the MLE results in part (b), we integrate the normal pdf from 215 to infinity (by computer) to yield approximately 0.2. Exercise 4.1.6. Show that the estimate of the pmf in expression (4.1.9) is an unbiased estimate. Find the variance of the estimator also. Solution: The expression (4.1.9) is n p̂(aj ) = where 1X Ij (Xi ), n i=1 Ij (Xi ) = 1 X i = aj , 0 Xi = 6 aj and the sample space is D = {a1 , . . . , an }. Now, we have # " n n n 1X 1X 1X 1 Ij (Xi ) = E[Ij (Xi )] = P (Xi = aj ) = nP (Xi = aj ) = p(aj ). E n i=1 n i=1 n i=1 n Hence, p̂(aj ) is an unbiased estimator of p(aj ). Now, the variance of the estimator is ! n 1X 1 1 Ij (Xi ) = V ar(Ij (Xi )) = p(aj ) 1 − p(aj ) . V ar(p̂(aj )) = V ar n i=1 n n 3 Exercise 4.4.5. Let Y1 < Y2 < Y3 < Y4 be the order statistics of a random sample of size 4 from the distribution having pdf f (x) = e−x , 0 < x < ∞, zero elsewhere. Find P (Y4 ≥ 3). Solution: Let X be a random variable with pdf f (x). We have P (Y4 ≥ 3) = 1 − P (Y4 < 3) 4 = 1 − P (X < 3) Z 3 4 −x =1− e dx 0 4 = 1 − 1 − e−3 . Exercise 4.4.9. Let Y1 < Y2 < . . . < Yn be the order statistics of a random sample of size n from a distribution with pdf f (x) = 1, 0 < x < 1, zero elsewhere. Show that the kth order statistic Yk has a beta pdf with parameters α = k and β = n − k + 1. Solution: Let gk (yk ) denote the pdf of Yk , and F (yk ) the cdf of Yk . Then, using the formula for kth order statistic pdf, we have n! [F (yk )]k−1 [1 − F (yk )]n−k f (yk ) 0 < yk < 1 (k−1)!(n−k)! gk (yk ) = 0 elsewhere k−1 n! n−k y (1 − yk ) 0 < yk < 1 (k−1)!(n−k)! k = 0 elsewhere ( Γ(k+n−k+1) k−1 y (1 − yk )n−k+1−1 0 < yk < 1 Γ(k)Γ(n−k+1) k = 0 elsewhere ( Γ(α+β) α−1 yk (1 − yk )β−1 0 < yk < 1 Γ(α)Γ(β) = . 0 elsewhere Hence, the kth order statistic Yk has a beta pdf with parameters α = k and β = n − k + 1. Exercise 5.1.2. Let the random variable Yn have a distribution that is b(n, p). (a) Prove that Yn /n converges in probability to p. This result is one form of the weak law of large numbers. Proof: Let {Xn } be a sequence of Bernoulli random variables P with parameter p. So, E[Xi ] = p, V ar(Xi ) = p(1 − p) < ∞ for all i ∈ N. Then, ni=1 Xi = Yn . Hence, by the weak law of large numbers, we have that n Yn 1X P = X̄n = Xi − → E[Xi ] = p. n n i=1 4 (b) Prove that 1 − Yn /n converges in probability to 1 − p. Proof: By part (a), we have P (|1 − Yn /n − (1 − p)| > ) = P (|Yn /n − p| > ) −−−→ 0. n→∞ P Hence, 1 − Yn /n − → 1 − p. (c) Prove that (Yn /n)(1 − Yn /n) converges in probability to p(1 − p). P P Proof: Using the theorem in the section, since Yn /n − → p and (1 − Yn /n) − → (1 − p), P we have that Yn /n(1 − Yn /n) − → p(1 − p). Exercise 5.1.3. Let Wn denote a random variable with mean µ and variance b/np , where p > 0, µ, and b are constants (not functions of n). Prove that Wn converges in probability to µ. Proof: Using Chebyshev’s inequality, we have P (|Wn − µ| ≥ ) ≤ b/np V ar(Wn ) = −−−→ 0. 2 2 n→∞ P → µ. Hence, by definition of convergence in probability, Wn − Exercise 5.2.1. Let X̄n denote the mean of a random sample of size n from a distribution that is N (µ, σ 2 ). Find the limiting distribution of X̄n . Solution: We look at the limiting behavior of the cdf of X̄n . Since the sum of normal random variables is normal with sum of parameters, we have FX̄n (t) = P (X1 + . . . Xn ≤ nt) Z nt (x − nµ)2 1 √ exp − = dx 2nσ 2 2πnσ 2 −∞ 2 x−nµ Z nt √ nσ 1 √ exp − dx. = 2 2πnσ 2 −∞ Now, change variables as x − nµ u= √ nσ 1 =⇒ du = √ dx, nσ 5 which yields Z nt −∞ 2 x−nµ Z √n(t−µ)/σ √ nσ 1 √ exp − dx = 2 2πnσ 2 −∞ 0 1/2 −−−→ n→∞ 1 1 2 √ e−u /2 du 2π t<µ t=µ . t>µ Now, note that F (t) := 0 t<µ 1 t≥µ is a cdf and limn→∞ FX̄n (t) = F (t) at every continuity point t of F (t). Hence, FX̄n converges in distribution to a random variable that has a degenerate distribution at t = µ. 6